20 December 2024 in Ad Hoc Networks
27 April 2026 in IEEE Internet of Things Journal
With the rapid advancement of intelligent and connected vehicles, in-vehicle networks face increasing demands for high-capacity data processing and transmission. Traditional static scheduling methods in Time Sensitive Network (TSN) can no longer satisfy the stringent real-time requirements of dynamic environments. The scheduling of time-sensitive traffic is typically addressed through offline computation, producing a mathematically constrained schedule that is subsequently deployed to end nodes or switches for configuration. However, such static deployment fails to accommodate dynamic scenarios. Consequently, this study proposes a dynamic incremental scheduling framework for TSN based on Deep Reinforcement Learning (DRL). The proposed framework integrates static global optimization with rapid dynamic adaptability. In the static phase, a Deep Q-Network (DQN) learns the priority ordering of time-sensitive flow scheduling under the given network state, thereby guiding a heuristic scheduler to efficiently allocate link resources. In the dynamic phase, an incremental scheduling strategy locally reconfigures resources for newly introduced flows without affecting existing schedules, thereby enhancing responsiveness and feasibility. Furthermore, this study proposes a load-balancing-guided routing set generation method that jointly optimizes path hop count and link load, enabling coordinated scheduling and routing. Comparative analysis under various network load scenarios demonstrates that the proposed method significantly reduces the response time of dynamic flow scheduling by 35% compared to the Satisfiability Modulo Theories (SMT) algorithm, while maintaining scheduling feasibility.
4 February 2026 in IEEE Transactions on Vehicular Technology
In modern automotive zonal electrical/electronic architectures, Time-Sensitive Networking (TSN) serves as the communication backbone, while Controller Area Network (CAN) operates as regional communication branches. The cross-domain communication requirements among CAN nodes distributed across different zones make CAN-TSN scheduling critically important. However, existing scheduling methods struggle to simultaneously meet CAN messages' real-time constraints and achieve efficient TSN bandwidth utilization, especially under high-load conditions with multiple CAN domains. To address these challenges, this paper introduces a comprehensive system model for CAN-TSN scheduling and, for the first time, establishes an aggregation theorem, providing a rigorous theoretical foundation for efficient CAN frame aggregation based on the AVTP protocol. Leveraging this theoretical result, a Constructive Heuristic Algorithm for AVTP-based CAN-TSN scheduling (CHAC) is proposed to produce initial high-quality scheduling solutions. Further optimization is conducted using a Hybrid Iterative Greedy Algorithm (HIGA), which integrates tabu list strategies and the Metropolis acceptance criterion to explore and refine solutions effectively. Simulation studies and prototype experiments conducted under heavy traffic scenarios demonstrate the effectiveness of the proposed methods. Compared to state-of-the-art approaches, the TSN link load is reduced by 41.7%, gateway CPU utilization decreases by 60.3%, and the CAN frame reception rate increases to 93.6%. These outcomes highlight the significant contribution of this research in advancing CAN-TSN scheduling performance, particularly under high-load scenarios, by achieving superior bandwidth utilization and robust real-time communication performance.